| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 56.54871101379395, | |
| "base_token_generation_latency_async": 57.30791725218296, | |
| "base_token_generation_throughput_sync": 0.017683869040906516, | |
| "base_token_generation_throughput_async": 0.01744959593627368, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 54.76382751464844, | |
| "base_inference_latency_async": 54.30099964141846, | |
| "base_inference_throughput_sync": 0.018260228427834344, | |
| "base_inference_throughput_async": 0.018415867232713764, | |
| "base_inference_CO2_emissions": null, | |
| "base_inference_energy_consumption": null, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 43.855078887939456, | |
| "smashed_token_generation_latency_async": 43.867955915629864, | |
| "smashed_token_generation_throughput_sync": 0.022802376038479982, | |
| "smashed_token_generation_throughput_async": 0.022795682614509663, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 51.30598335266113, | |
| "smashed_inference_latency_async": 45.325446128845215, | |
| "smashed_inference_throughput_sync": 0.019490904074994833, | |
| "smashed_inference_throughput_async": 0.022062662045450838, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |